Why a myth outlives its debunking
A myth that survives usually has three things going for it. It is cheap to believe, because acting on it costs a checkbox rather than a rewrite. It is impossible to disprove from your own data, because a site that ranks after doing it cannot show what would have happened otherwise. And it is riding on a real correlation — something true sits next to it, close enough to lend it credibility.
That last part is why flat contradiction rarely kills one. Longer pages really do often outrank shorter ones; it is just that the length is a symptom of covering the subject, not the cause of the ranking. Sites with older domains really do often rank better; they have had longer to earn links and build a record. Pull the correlation apart from the claim and most of these stop being arguable.
On-page claims with nothing behind them
These come up in nearly every audit conversation, and each one has been addressed directly by Google rather than merely doubted by practitioners.
| Claim | What the documentation says |
|---|---|
| The keywords meta tag helps | Unused for ranking, and stated as such since 2009 |
| There is an ideal word count | No target exists, minimum or maximum |
| Keyword density has a right number | No such number; stuffing is named as spam |
| A page must have exactly one H1 | Multiple headings of any level are fine |
| Headings must run in strict order | Out-of-order levels cause no Search problem |
| "LSI keywords" need to be added | Not a thing Google's systems use |
The heading ones are worth a second look, because they are where audit tools do the most damage. A grader that deducts points for two H1s is reporting its own rule, not the engine's. Headings should describe the sections under them because that helps a reader scan and helps a machine understand structure — which is a reason to write them well, not a reason to chase a number.
The duplicate content penalty
There isn't one. Google's position is that the same content under several URLs is not grounds for action against a site, unless the duplication is deliberately deceptive. What normally happens instead is consolidation: the engine picks one URL as canonical and folds the signals of the others into it.
That does not make duplication free. It costs you in two ways, both of which are housekeeping problems with housekeeping fixes. Signals that should have accumulated on one page get spread across several near-copies. And crawl effort goes into fetching pages that add nothing, which matters on a large site and is invisible on a small one.
The distinction is worth holding onto because it changes the fix entirely. A penalty would call for removal and a reconsideration request. Consolidation calls for declaring the canonical yourself instead of leaving the choice to the engine — a tag, not an apology.
Links, domains and money
| Claim | What the documentation says |
|---|---|
| Buying Google Ads lifts organic rankings | Explicitly denied; the two systems are separate |
| Older domains rank better for being older | Domain age on its own does nothing |
| A keyword in the domain gives a boost | Very little effect, and reduced deliberately in 2012 |
| nofollow can channel PageRank internally | Sculpting stopped working in 2009; now a hint |
| Sites must be submitted to search engines | Not required; discovery happens by crawling links |
| Analytics data feeds the ranking systems | Google states it does not use Analytics for ranking |
Two of these deserve expanding. nofollow was reinterpreted in 2019 as a hint rather than a directive, for both crawling and ranking — so the old sculpting trick fails twice over: it does not conserve anything, and the link may be followed anyway. And the metric people most want to be a ranking factor, bounce rate, is not one; the engine has no reliable access to it and says so.
The new-site question is the one worth answering honestly rather than dismissing. Google denies operating a formal sandbox that holds new domains back. What is true is that a new site has almost no history for the systems to reason about — no links, no record of satisfying anyone — and gathering that takes time. The waiting is real. The mechanism people imagine is not.
The crop that arrived with AI
Generative search produced a fresh set of tactics inside a year, and they follow the pattern exactly: cheap to adopt, impossible to disprove locally, and riding on the real observation that AI answers behave differently from ten blue links.
- A file declaring your content to language models. Google has said Search does not use llms.txt, and the comparison drawn was to the keywords meta tag — a file you control, asserting things about yourself, with no way to verify them.
- Content pre-chopped into machine-sized pieces. Retrieval systems do their own segmentation; writing in fragments to pre-empt it costs you the reader who has to read the fragments.
- A second version of the page written for machines. This is cloaking with a modern justification, and it is the same spam-policy violation it was when the second version was written for crawlers.
What actually governs whether a page can appear in an AI answer is unglamorous and already documented: the page has to be in the index and eligible to show a snippet. That is the whole technical requirement. Everything past it is the same work as before — being the better answer to the question, in a page an engine can fetch, render and store.
Testing a claim before you spend on it
Most of these can be settled in about five minutes, and the habit generalises to claims that have not been written down yet.
- Find the primary source. A blog post citing a blog post citing a conference talk is not one; the engine's own documentation or a named statement from it is.
- Check what the claim is actually about. Eligibility for a feature, ranking, and crawling are three different outcomes, and advice routinely promises one while citing evidence for another.
- Check the date. Search changes, and a statement that was accurate in 2015 may have been superseded twice since.
- Name the correlation it is riding on. If the tactic is a symptom of something real, do the real thing instead and the symptom follows.
- Price being wrong. Cheap and harmless can stay on a hunch; anything that costs a sprint needs the source you found in step one.